Nvidia sustainability chief says AI could cut emissions as power use rises

In short: Nvidia’s head of sustainability says AI is getting far more energy-efficient and may help other industries use less energy, even as AI increases electricity demand.

What's going on

Josh Parker, Nvidia’s head of sustainability, is arguing that AI systems are becoming much more energy-efficient over time. He says this matters because AI can be used to help factories, buildings, transportation, and farms run more efficiently, which could lower carbon emissions.

Nvidia says its accelerated computing systems, which use GPUs (special chips often used for graphics and AI), can be up to 20 times more energy-efficient than using CPUs alone for AI work. Parker has also pointed to a comparison where AI “inference” efficiency improved by about 45,000 times in eight years. Inference is when an AI model is actually used, like when a chatbot answers a question, rather than being trained.

Parker’s main point is about system-level savings in the future, not a near-term drop in AI’s own electricity use. He has cited examples like “digital twins,” which are computer copies of real-world factories or buildings (like a flight simulator, but for an industrial site). In one case he cited, the approach achieved a 10% energy reduction, and another manufacturing example was forecast to cut lifetime energy use by 30%.

What to watch

The argument is contested because data centers, which are large buildings full of computers that run AI, are using more electricity as AI demand grows. Some critics say the efficiency gains may not offset AI’s total footprint quickly enough, and some reporting has accused Nvidia of greenwashing. Parker has also promoted “flexible data centers” that can adjust power use when the electric grid is under stress, and Nvidia says it has matched 100% of its global electricity use with clean energy for a second year.

Source: NYTimes